Amazon EC2 (Elastic Compute Cloud) is one of the most frequently tested services on the AWS Certified Cloud Practitioner exam. While the CLF-C02 does not ask you to memorise specific instance names like m5.xlarge or c6g.2xlarge, it absolutely expects you to understand the families of EC2 instances, what workloads each family is optimised for, and — crucially — how to choose the right one for a given scenario.
This guide walks through every instance family you need to know, explains the thinking behind them, and ties it all together with exam-style reasoning you can apply on test day.
Why Instance Families Exist
When AWS designed EC2, it faced a fundamental challenge: different applications have radically different appetites for CPU, memory, network throughput, and storage. A video-transcoding job needs raw processing power; a large in-memory database needs enormous RAM; a machine-learning training job needs specialised GPU hardware.
Rather than building one-size-fits-all servers (which would be expensive and inefficient for most use cases), AWS created instance families — groups of instances tuned for specific performance profiles. Understanding this design philosophy is the key to answering scenario questions correctly, because the exam almost never asks "what is an M-series instance?" — it asks "which instance type is best suited for a web application with balanced CPU and memory needs?"
The Five EC2 Instance Families
The CLF-C02 exam guide groups EC2 instance types into five families. Here is each one in plain language.
1. General Purpose
Use when: You need a balanced mix of compute, memory, and networking and do not have a dominant bottleneck.
General Purpose instances are the workhorses of EC2. They are designed for applications where no single resource (CPU, memory, or network) is the obvious constraint. Think of them as the "default" choice.
Typical workloads: - Web servers and small-to-medium web applications - Application servers - Development and test environments - Code repositories
Exam tip: If a scenario describes a workload as "general web application," "development environment," or "balanced," General Purpose is almost certainly the right answer.
2. Compute Optimized
Use when: Your application is CPU-heavy — it needs more processing power relative to memory.
Compute Optimized instances provide a higher ratio of CPU to memory. They are purpose-built for workloads that spend most of their time doing computation.
Typical workloads: - Batch processing and high-performance computing (HPC) - Scientific modelling - Dedicated gaming servers - Video encoding - CPU-based machine learning inference
Exam tip: The word "compute" in the name is your clue. If a scenario mentions "high-performance computing," "batch processing," or "CPU-intensive," think Compute Optimized.
3. Memory Optimized
Use when: Your application needs to work with very large datasets entirely in RAM, because accessing memory is dramatically faster than accessing disk or even a fast SSD.
Memory Optimized instances offer much higher RAM-to-CPU ratios than General Purpose. They are designed for workloads where having more memory directly translates to better performance.
Typical workloads: - In-memory databases (e.g., Redis, Memcached) - Real-time processing of large unstructured data - High-performance relational databases with large working sets - SAP HANA and other enterprise in-memory applications
Exam tip: Keywords to watch for: "in-memory," "large database," "real-time big data processing," or any mention of keeping a huge dataset in RAM.
4. Storage Optimized
Use when: Your workload requires very high, sequential read/write access to large datasets on local storage — specifically when you need high IOPS (input/output operations per second).
Storage Optimized instances are equipped with high-speed NVMe SSDs and are tuned for I/O-heavy workloads. They are not about how much storage you can attach — they are about how fast you can read and write.
Typical workloads: - Data warehousing - Distributed file systems (e.g., Hadoop) - High-frequency transaction processing databases - Online transaction processing (OLTP) systems requiring high I/O - Elasticsearch / OpenSearch clusters
Exam tip: Look for phrases like "high I/O," "high IOPS," "sequential read/write," or "distributed file system." Storage Optimized is the answer when throughput to local disk is the bottleneck.
5. Accelerated Computing
Use when: Your workload can benefit from hardware accelerators — GPU or FPGA — rather than relying solely on general-purpose CPUs.
Accelerated Computing instances attach specialised processors that are orders of magnitude faster than CPUs for specific tasks like floating-point calculations, graphics processing, or data pattern matching.
Typical workloads: - Machine learning training (especially deep learning) - High-performance graphics and game streaming - Genomics research - Financial modelling and simulations - Video transcoding at scale
Exam tip: "GPU," "machine learning training," "deep learning," "graphics rendering," and "floating-point operations" all point to Accelerated Computing.
Not sure where you stand?
Take the free 10-question CLF-C02 diagnostic and get an instant score, your weakest domain, and a study plan.
Take the free diagnosticHow to Think About Instance Selection on the Exam
The CLF-C02 is a scenario-based exam. Rather than testing definitions, it presents a situation and asks which instance family fits best. Here is a simple decision framework:
- Is there a hardware accelerator mentioned (GPU, FPGA)? → Accelerated Computing
- Is the bottleneck memory / RAM? → Memory Optimized
- Is the bottleneck disk I/O / IOPS? → Storage Optimized
- Is the bottleneck raw CPU? → Compute Optimized
- No clear bottleneck — balanced workload? → General Purpose
Practice applying this framework with exam-style questions. Our practice questions include scenario-based EC2 questions that help you build this instinct quickly.
EC2 Instance Types and the Pricing Models
Understanding instance families is only half the picture. The CLF-C02 also tests how you pay for EC2. The pricing model you choose is independent of the instance family — you can run a Memory Optimized instance on On-Demand pricing, a Spot Instance, or a Reserved Instance.
Here is a quick recap of pricing models (covered in Domain 4):
- On-Demand: Pay by the second or hour, no upfront commitment. Best for unpredictable workloads.
- Spot Instances: Bid on unused EC2 capacity at significant discounts — up to around 90% off On-Demand pricing. AWS can reclaim the instance with a two-minute warning, so these suit fault-tolerant, interruptible workloads like batch jobs.
- Reserved Instances (Standard): Commit to a 1- or 3-year term for the highest discounts. Best for steady, predictable workloads.
- Reserved Instances (Convertible): Similar commitment, but you can change instance attributes. Slightly lower discount in exchange for flexibility.
- Savings Plans: Flexible commitment-based discounts that apply across instance families (Compute Savings Plans also cover Lambda and Fargate).
- Dedicated Hosts: A physical server dedicated to your use — typically chosen for compliance or bring-your-own-licence (BYOL) requirements.
A common exam question pairs an instance family with the right pricing model for a scenario. For example: "A data science team needs GPU instances to train models overnight on weekends only. Which pricing model minimises cost?" The answer would point toward Spot Instances, since training jobs are often interruptible and the workload is not continuous.
Where EC2 Instance Types Fit in the CLF-C02 Domains
EC2 falls squarely in Domain 3: Cloud Technology and Services, which carries the heaviest weighting on the exam at 34%. EC2 instance families, along with concepts like serverless compute (Lambda), managed databases (RDS, DynamoDB), and storage (S3), make up a significant slice of Domain 3 content.
If you are not sure how well you understand Domain 3 relative to your other weak spots, our free diagnostic can help you identify exactly where to focus your study time.
Quick-Reference Summary
| Family | Optimised For | Typical Exam Keywords |
|---|---|---|
| General Purpose | Balanced CPU/memory/network | Web server, app server, dev/test |
| Compute Optimized | High CPU-to-memory ratio | HPC, batch processing, gaming server |
| Memory Optimized | High RAM-to-CPU ratio | In-memory DB, real-time big data |
| Storage Optimized | High local disk I/O / IOPS | Data warehouse, OLTP, distributed FS |
| Accelerated Computing | GPU / FPGA hardware | ML training, graphics, deep learning |
FAQ
Do I need to memorise specific instance names like m5.xlarge for the CLF-C02?
No. The CLF-C02 tests your understanding of instance families and which workloads they suit, not individual instance names or sizes. Focus on recognising the five families and matching them to scenario descriptions.
What is the difference between Storage Optimized and Memory Optimized instances?
Memory Optimized instances are built for workloads that need a very large amount of RAM — the data lives in memory. Storage Optimized instances are built for workloads that need extremely fast reads and writes to local disk storage, measured in IOPS. If a question mentions RAM or in-memory processing, think Memory Optimized. If it mentions I/O, IOPS, or local disk throughput, think Storage Optimized.
Can I use Spot Instances with any EC2 instance family?
Yes. Spot pricing is a billing model and is available across all instance families, not limited to a specific family. The key constraint is that AWS can reclaim Spot Instances with two minutes' notice, so they are only appropriate for interruptible workloads.
How much of the CLF-C02 exam covers EC2?
The exam does not publish a per-service breakdown. EC2 falls under Domain 3 (Cloud Technology and Services), which is the largest domain at 34% of the exam. EC2 — including instance types, pricing models, and related concepts — is one of the most tested services within that domain.
What instance family would you use for machine learning training?
Accelerated Computing. ML training — especially deep learning — benefits enormously from GPU acceleration. The exam will often pair "machine learning training" or "deep learning" with Accelerated Computing as the correct answer.
Is General Purpose always the wrong answer for specialised workloads?
Not necessarily, but it is rarely the best answer when a scenario describes a clear bottleneck. If a question says "a company needs to run a web application with no unusual performance requirements," General Purpose is correct. If any specific resource constraint is mentioned, a specialised family will almost always be a better fit.